Stream a batch of translations

With batch translation already covered, streaming lets Bergamot return each translation as it's ready, instead of everything at once.

stream: true turns the same request into a duplex session. Rather than holding the batch until its last entry finishes, the engine hands over each translation as it completes, so a long list shows steady progress.

We can switch to streaming by setting stream: true:

const result = translate({
  modelId,
  text: texts,
  modelType: "nmtcpp-translation",
  stream: true,
});

A streaming request doesn't resolve result.translations. Iterate result.tokenStream instead, and track an index yourself so you can match each translation to the entry that produced it:

let index = 0;
for await (const translation of result.tokenStream) {
  console.log(`${index + 1}. ${texts[index]} -> "${translation}"`);
  index++;
}
console.log(`Translated ${index} texts`);

The stream produces the translated string, and texts[index] gives you its source.

Note: despite the name, tokenStream yields one finished translation per entry here, not one token at a time.